01 The thing I wanted didn't exist.
The Samantha from Her. Not to fall in love with. To challenge me. Something that argues format-accurately, holds the clock, takes my points, and writes a real judge ballot when the round ends.
Prep was always the bottleneck. You can flow your own case all night. You can read until your eyes burn. None of it simulates the moment an opponent stands up and says your second contention contradicts your framework. For that you need someone on the other side, and most nights you don't have one. The people who would take a round with you are tired of running your drills.
02 Why I think this matters.
Arguing out loud, against a clock, against someone who wants to win. That's where thinking actually sharpens. Not just for debaters. Lawyers prep crosses, founders rehearse Q&A, anyone who has to be coherent under pressure benefits from a partner who pushes back instead of nodding.
03 What it doesn't replace.
This does not replace a coach. A coach reads you, builds your judgment, names the habit you can't see in yourself. DebateAI does the one thing a coach can't be on call for: as many reps as you want, any format, any hour. Coaching is scarce. Practice shouldn't be.
04 Why this exists, in order.
05 Open problems I'd take help on.
- Realistic interruption handling. Voice round needs to feel like a real POI exchange — sub-200ms barge-in, judge of when to yield, judge of when to refuse.
- AI judging consistency. The same motion + same speeches should not produce a different ballot brain-to-brain. Multi-brain consensus is a stopgap, not a solution.
- Multilingual debate formats. Hindi-medium Asian Parli, Mandarin WSDC, Spanish college rounds. The infra is there; the format-specific register isn't.
- Live tournament infrastructure. AI prelims into human elims, bracketing, panel judging, tab software in the browser.
- Memory between rounds. The AI should remember how you argued last week and play you differently this week. Style fingerprint exists; round-to-round adaptation is thin.
06 What I believe.
Adversarial reasoning matters. Oral argument is undervalued. AI should pressure-test thinking, not flatter it. The product is debate-first because that's where the moat is and because that's the room I came up in. Everything else, eventually, runs on the same engine.